Predictive Analytics for Sand Production: a Machine Learning Study from the Niger Delta

Sand production remains a critical challenge in petroleum extraction, particularly in weakly consolidated sandstone reservoirs. The presence of sand in produced fluids can lead to erosion of equipment, diminished well productivity, and an increase in operational costs. However, conventional sand management techniques, like gravel packing and chemical consolidation, do not perform well due to the ambiguous interactions between reservoir geomechanics, fluid properties, and operational conditions. To address this challenge, this study leverages machine learning techniques to develop a predictive model for sand production, integrating key parameters to enhance accuracy and adaptability in real-world applications.

Models are trained and validated using Dataset from 7 oil fields in Niger Delta comprised of key reservoir and fluid properties such as water cut, Gas-oil ratio (GOR), Bean size, oil rate, Tubing-Head Pressure (THP), permeability etc. The flow bottom-hole pressure (FBHP), gas-oil ratio (GOR) and permeability emerge as the top three contributors to sand production. Three machine learning algorithms—Random Forest, XGBoost, and Linear Regression—were implemented, and their performance was evaluated using accuracy, precision, recall, R2 score, and root mean squared error (RMSE).

Using the developed predictive model, the R2 score showed a high degree of accuracy when forecasting sand production trends, with a value of R2 = 0.9547. The Random Forest and XGBoost algorithms performed better than most of the other models tested in terms of their capacity to explain the nonlinear relationship between sand production. This proved itself to be robust at predicting sanding onset and severity and thus has value in aiding proactive sand management in oilfield operations through model validation with historical field data.

Thus, this study provides a data-driven framework to optimize the sand management strategies, minimize the operational risks and enhance the decision-making in petroleum production.